MAPPING URBAN PERCEPTION: HOW DO WE KNOW WHERE WE ARE?
BY
SHAN HE
SUBMITTED TO THE
DEPARTMENT OF ARCHITECTURE IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE IN ARCHITECTURE STUDIES
MASACHES N6TEJ'
AT THE
MASSACHUSETTS NNTE
OF TECHNOLOGY
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
JUL 0 1 2014
JUNE 2014
LIBRARIES
©2014 Massachusetts Institute of Technology. All rights reserved.
Signature redacted
Signature of Author....
................................
Department of Architecture
May 27, 2014
Certified by..................................
nature redacted
%0
V Taehi
Ass ociate Professor of De * n a d
Nagakura
putation
s pervisor
Signature redacted
C ertified by.............................................
-- a lo Ratti
Associate Professor of the Practice
Director, Senseable City Lab
Thesis Co-supervisor
Accepted by...........................................
Signature redacted,-
Takehikd Nagakura
sociate Professor of Design and Computation
Chair of the Department Committee on Graduate Students
Thesis supervisor:
Takehiko Nagakura
Associate Professor of Design and Computation
Carlo Ratti
Associate Professor of the Practice
Director, Senseable City Lab
Thesis reader:
Federico Casalegno
Associate Professor of the Practice
Director, Mobile Experience Lab
Mapping Urban Perception: How Do We Know Where We Are?
by
Shan He
Submitted to the Department of Architecture on
May 22, 2014 in Partial fulfillment of the
Requirements for the Degree of Master of Science in Architecture Studies
ABSTRACT
How do we remember urban space? How can we measure what is remembered? This
thesis presents a new approach to study urban spatial perception in an efficient, automated,
and scalable way. It explores the use of novel tools developed for online surveys and data
visualization.
Previous studies in human spatial perception have used methods such as face-to-face
interviews, questionnaires, recognition tasks and surveys that ask subjects to draw sketch
maps. Those conventional methods produced significant urban studies such as the one by
Kevin Lynch (1960), but they are laborious, sensitive to the individuality of subjects, prone
to examiners' biases and conducted with a limited number of subjects. Their results are also
difficult to quantify. In contrast, the method developed here uses geo-tagged street views and
a web-based visual survey.
An online experiment conducted in this thesis collected 394 participants in 20 days who
were asked to guess the locations of street views from a familiar neighborhood. Results are
presented in the form of interactive visualizations. Analysis revealed that memory for exact
location of place improves with degree of interaction and proximity to center, rather than
number of encounters; memory for one location may vary dramatically between different
viewpoints. The results also suggest that the irregularity of urban structure doesn't prevent
the forming of strong mental images.
While this new method cannot completely replace face-to-face interviews, it
demonstrates the possibility of using available technology to scale visual surveys to hundreds
or even thousands of people and rapidly visualize the resulting data. It thus opens up new
possibilities for large-scale, fine-grained studies in urban perception.
Thesis supervisor:
Takehiko Nagakura
Associate Professor of Design and Computation
Carlo Ratti
Associate Professor of the Practice
Director, Senseable City Lab
TABLE OF CONTENTS
ACKNOWLEDGEMENTS
2
CHAPTER I INTRODUCTION
3
CHAPTER 2 FROM SPATIAL PERCEPTION TO THE COLLECTIVE
IMAGES OF THE CITY
Urban Perception -5
5
Mental Maps - 6
Units and Configurations of Mental Maps - 7
The Collective Images of the City - 11
Practical Applications - 12
CHAPTER 3 METHODOLOGY
13
Spatial Familiarity Evaluation - 13
Sketch Maps - 14
Surveys and Interviews - 15
Problems and Concerns - 16
Visual Surveys and Geo-tagged Images - 16
CHAPTER 4 THE EXPERIMENT
Tool Design - 19
19
Survey -21
Site - 21
Geo-tagged streetview images - 22
Participants - 24
Algorithm - 25
CHAPTER 5 RESULTS AND DISCUSSIONS
Most Familiar and Least Familiar - 28
Links and Nodes - 31
Knowledge Acquisition - 35
27
Modes of Interaction - 36
Discussion - 37
CHAPTER 6 VISUALIZATIONS
39
CHAPTER 7 CONCLUSIONS
45
BIBLIOGRAPHY
49
ACKNOWLEDGEMENTS
I would like to express my deep gratitude to Professor Takehiko Nagakura,
Professor Carlo Ratti and Professor Federico Casalegno for their guidance during the
development of this thesis. I would also like to thank Jonas Helfer for his technical support
during the development of the tool presented in this thesis.
I would like to give special thanks to Yuji Yoshimura and Stefan Seer who shared
precious suggestions with their expertise in this topic. Thanks to all the other researchers at
Senseable City Lab for their encouragement and positive input. Thanks to all the anonymous
participants who used this tool and provided valuable feedback.
Finally, I wish to thank my parents for their silent support of all my decisions all these
years. And to Jonas, again you made this happen.
Chapter 1
INTRODUCTION
An urban neighborhood is determined not only by geographical and
economic factors, but also by the images that its inhabitants and those of
other neighborhoods have of it.
-Chombart de Lauwe, Paris,essais de sociologie
In the five decades since Kevin Lynch's seminal work, The Image Of The City, scientists
from a variety of fields have conducted studies investigating the relationship between the
urban environment and the images its inhabitants have of it. The researchers focused on the
nature of mental maps, the process of acquiring spatial knowledge to form such maps, and
their roles in everyday spatial activities, such as navigation choice, way-finding behavior and
route learning (Golledge, 1997). A major challenge that remains is the externalization of
spatial knowledge: how to elicit cognitive data from people and what medium to use to
represent it. A wide range of techniques has been used to collect and analyze cognitive data.
Commonly used methods include verbal interviews, sketch maps, protocols, toy play,
matching tasks, proximity judgments, recognition tasks etc. (Golledge, 1997). Each of these
methods requires hours of interaction with subjects and a considerable amount of work to
compile and translate cognitive responses into quantitative data. The results are also sensitive
to idiosyncracies of subjects and their ability to master the testing media.
Beyond studying an individual's cognitive maps, city designers and planners are
primarily interested in the mental images that are shared by large numbers of city dwellers
(Lynch, 1964). By studying the collective mental maps and investigating areas of agreement,
designers and planners are able to examine which parts of the city are more likely to evoke
strong mental images. Such knowledge can provide guidance to design vivid urban
environments that facilitate rapid learning and easy orientation (Kitchin & Freundschuh,
2000). Cities are inhabited by millions of people and composed of billions of unique images.
To investigate how a variety of places in a large-scale urban environment are remembered by
their inhabitants, conventional methods with time-consuming interviews among a limited
number of subjects are evidently not sufficient. There is a need for methods that can be used
with a large number of subjects, and allow fine-grained observations in an efficient,
automated and scalable way.
4 1 Mapping Urban Perception: How Do We Know Where We Are?
Fortunately, today's dominant web technology enables crowd-sourced data collection.
Google's geo-tagged streetviews provide a large number of urban images. This thesis uses a
web-based visual survey in the form of a geo-guessing game to efficiently solicit people's
responses to how well they remember different urban images. In this study, participants were
asked to guess the locations of random streetviews from a familiar neighborhood by placing a
pin on a map. This method evaluates how well each urban image is remembered on the basis
of two spatial familiarity indicators: location identification and visual recognition. The data
collected were used to create interactive cartographic representations, and answer the
following questions: how are different types of spatial entities remembered by their
inhabitants; how is spatial memory acquired throughout time; and how do an individual's
daily activity patterns affect his comprehension of the urban environment.
The analysis concludes that the memory about the exact location of a place improves with
the degree of interaction and proximity to the center, rather than the number of encounters. It
reveals that memory for one location may vary dramatically when viewed from different
angles. The findings also suggest that the irregularity of urban structure doesn't prevent
forming of strong mental images, which challenges Lynch's statement that a regular city
structure is more likely to evoke strong mental images.
This thesis concludes that the method presented here is able to rapidly elicit spatial
knowledge among a large number of city dwellers, and allow us to conduct quantitative data
analysis. Moreover, this thesis proposes that the method can be used beyond the familiarity
evaluation of urban images. It can be adopted in future investigations including spatial
knowledge acquisition in an unfamiliar neighborhood, cognitive representation of spatial
relationships, and results of perceptional knowledge on human mobility behavior.
This thesis is organized as follows. Chapter 2 and Chapter 3 describe the precedent study
of mental images of the city and the methodology used. Chapter 4 presents the experiment.
Chapter 5 discusses the results. Chapter 6 presents data visualizations that were generated
during this thesis. Chapter 7 presents conclusions and offers suggestions for future research.
Chapter 2
FROM SPATIAL PERCEPTION TO THE COLLECTIVE IMAGES OF THE CITY
If we are aware of people's preferences for, perceptions of, and attitudes
toward different environments, then better matches between planning and
policy making and the felt needs of the populations, for whom plans are
being made, can be achieved.
-Gdrling, T., & Golledge, R. G. (1989)
In the scope of urban and city studies, the term "imageability" evaluates how well the city
spaces are understood and remembered by their inhabitants. The concept of "imageability"
was first established by Kevin Lynch (1964), an American urban planner, in his influential
book The Image Of The City. Lynch's study focused on how people perceive space and
navigate through the urban environment. He states that people navigate in a familiar urban
environment by means of mental maps-a memory representation of spatial information in
our minds, also referred to as cognitive maps (Lynch, 1964; Golledge, 1997). According to
Lynch (1964), a highly imageable urban environment has a high probability of evoking a
strong image among various observers. This chapter discusses the nature of human urban
perception, how we use mental maps to comprehend the complex urban environment and
how spatial information is obtained, stored and organized to form mental maps.
Urban Perception
Millions of people travel through the city everyday. What they choose to see and not to
see, what they like and do not like, what they remember and do not remember is of profound
interest to city planners and architects. The urban environment is a complex, formless and
fragmented system. Empirical studies suggest what people are more likely to store in their
memory-buildings with dominant shape and bright color, historical sites, intruding
signage-yet little has been done to systematically survey different aspects of the city and
investigate the probability of each to evoke strong mental images among a large number of
residents.
We choose to perceive urban environment for different purposes, such as looking for
places to stop and paying attention to the traffic lights. Appleyard (1973) in his influential
6 1 Mapping Urban Perception: How Do We Know Where We Are?
essay Notes on urban perception and knowledge proposes that people perceive the urban
environment in a categorical fashion. He categorized urban perception into three types:
operational, responsive and inferential. Operational perception is performed as a person uses
the city. People are more likely to remember the details of traffic circles and subway stations.
Responsive perception is passive perception. Buildings with bright color and distinctive
shapes are more likely to catch people's eyes. This type of perception is extensively
discussed in Lynch's imageability theory. The third type of perception is inferential, which
emphasizes actions of synthesis and generalization when new experiences are matched with
old experiences (Appleyard, 1973). People are more likely to remember the Bank of America
branch when moving to a new neighborhood, if they have seen a similar looking branch
previously.
Clearly, the categorization proposed by Appleyard is more theoretical than practical,
because it is almost impossible to isolate each type to examine its role in obtaining a specific
type of spatial knowledge. Appleyard himself also stated that our spatial knowledge is a
complex collection of varyingly perceived items, qualities, and events. It is a multi-modal
representation of the city (Appleyard, 1973). His theory was the first attempt to connect
urban spatial perception with human perceptual and cognitive processes and provides a
theoretical framework for studying the inhabitants' images of the city.
Mental Maps
Have you ever lost your way in your own neighborhood? Have you ever found yourself
losing orientation immediately after coming out of a different exit of the subway stop you
take every day? A bear can find its way to honey using smell and its instinct alone; a dog
abandoned by its owner thousands of miles from home eventually finds its way back after
three months. Some animals even possess specialized senses and faculties to navigate
through familiar or unfamiliar environments, such as magnetic field sensitivity or the ability
to navigate using the sun as a reference point (Ellard 2010). Humans, on the other hand,
while superior in our cognitive abilities, always seem to perform badly when it comes to
figuring out where we are. It's easy to see how bad we are at dealing with our relationship
with space simply by counting the number of apps we have on our smartphones to help us
find our way.
From Spatial Perception To The Collective Images Of The City 1 7
While other animals possess prodigious talents for memorizing their paths or the subtle
hints in the natural environment, human beings comprehend the urban environment by
following a somewhat different course (Ellard 2010). Cities are man-made large-scale
environments, filled with salient and familiar views, and composed of spatial objects with
distinctive forms (e.g., buildings, roads, squares, bridges etc.). Humans, survivors of this
complex environment, have to sense, store, record, and organize bits of spatial information
into exclusive knowledge sets with the task of living (R. G. Golledge, 1997). The product of
this living knowledge process is an abstract organization of spaces in our minds.
Psychologists often refer to this organization of space as mental maps or cognitive maps. To
perform everyday navigation tasks such as going to work, picking up dry cleaning or
describing how to get to one's home, people rely on their mental maps of the city.
The mental maps we construct in our minds are highly schematized, and often bear little
resemblance to the real environment. This is because our mind tends to simplify visual
patterns. We straighten curves, simplify geometry, and reduce complex spaces to a simple
collection of basic shapes (cited in Ellard, 2010). Barbara Tversky, a cognitive psychologist
at Stanford University, conducted groundbreaking studies on cognitive mapping, especially
the differences between mental maps and the physical spaces they represent. One important
finding of Tversky's study is that our minds tend to blur distance and direction but treat
topological relations with great clarity (Ellard, 2010). When asked to imagine how to travel
from point A to point D, we tend to first think about how to get to B, then C, and finally D.
Although we have a good understanding of how places are connected, we don't seem to be
able to accurately recall either the distance between places or their directional relationship
(imagine how many times we find ourselves saying: "I know how to get to A, but I am not
sure if it's the shortest route").
Units and Configurations of Mental Maps
Mental maps are formed through a way-finding process, just as a mouse's spatial memory
is formed through its searching for cheese in a maze. The study of mental maps is concerned
with how we comprehend physical space by storing, organizing and positioning spatial
information internally. Despite the different terminology that is used, it is commonly agreed
that mental maps primarily represent spatial relationships and contain some map-like
qualities (Evans, 1980). They consist of the topological relationships of spaces, rather than
8 1 Mapping Urban Perception: How Do We Know Where We Are?
their absolute coordinates and distances. This section discusses the basic units perceived from
the environment and the schemata that are applied in the process of forming mental maps.
From the environment, we obtain spatial knowledge such as where the supermarket is,
how far the next intersection is, which road passes by our home etc., by living in it and
traveling through it. The basic elements of mental maps are environmental units that are most
familiar to us (R. Golledge, 1992). However, different hypotheses exist in cognitive
psychology about how these environmental units are recorded, stored and represented
internally. Neuroscientists believe that there is a certain brain mechanism that connects
specific neurons to encode the perceived spatial information; most psychologists believe that
spatial information stored in our mind exists in some type of psychological space whose
metric system is unknown (Ellard 2010). In their perspective, the elements in such space
maintain rough, isomorphic correspondence to the physical world often by their functional
property, rather than physical forms (Evans, 1980).
In urban perception and cognition studies, the most accepted view is that that the
environment is stored in the format of symbolic representations using "Euclidean metric",
attached with abstract labels (Golledge 1999, Evans, 1980). In fact, much of human spatial
knowledge is traditionally represented as a Euclidean metric with basic elements including
points, lines, areas and surfaces (R. G. Golledge, 1999). Some researchers believe that the
maps we sketch are the best representation of the properties of space that inhabit our minds
(Ellard 2010). One example is the sketch maps that Lynch asked the local people to draw in
order to illustrate their memories of the city. Lynch stated that mental maps reflect basic
elements of the physical city form (as cited in Evans, 1980). He suggested five basic
elements as primitive elements of mental maps: paths, edges, districts, nodes and landmarks
(Lynch, 1964).
Paths are routes people travel along, such as streets, roads and sidewalks. Edges
are usually boundaries, such as buildings and riverside. Districts are large areas of
the urban space that are usually referred to be the same identity, such as
downtown, SOHO and the upper east side of Manhattan. Nodes are locations that
are usually referred as strategic focus points, a busy intersection or a public
square are such examples. Landmarks by definition are spatial elements that are
frequently identified as reference points.
From Spatial Perception To The Collective Images Of The City
Path
Edge
District
Node
|
Landmark
Figure 1.1 Lynch's five basic elements of the mental maps
Source: Lynch, 1964
Lynch asked local people from Boston to draw sketch maps of downtown Boston as they
remembered it. Based on all the drawings, Lynch constructed a map of downtown Boston
representing the structure of its spatial elements reflected in people's memories, and how
well they were remembered. He used simple geometry to represent spatial elements: lines
represent paths and edges (boundaries); circles represent nodes (intersections); areas
represent districts; and triangles represent landmarks. The thickness of the lines, darkness of
the circles and areas represent the strength of mental images evoked by that element.
I-.
K-
k,
0
PATH
NGCE
EDCE
DSTRICT
LANttIARK
over 75% freqvwey
.
50-75%
25-501
---
V
V
V
Figure 1.2 The view of Boston constructed in Lynch's study
Source: Lynch, 1964
9
10
|
Mapping Urban Perception: How Do We Know Where We Are?
Lynch's five elements were derived primarily from the shapes that repeatedly used in
people's hand sketches and coded with respect to the physical forms of the actual elements.
He proposed that paths were the structural feature that connected all the others elements. He
also stated that paths are what people primarily use to navigate through the urban
environment (Lynch, 1964). Golledge, on the other hand, suggested that landmarks were the
organizing features in a way-finding context. He stated that landmarks acted as anchor points
for organizing other spatial information into a layout (R. G. Golledge, 1999). The "anchor
point" theory proposed by Golledge and Spector (1978) suggested that the complex urban
forms stored in our memory in the form of a linked-node configuration. Nodes are spatial
reference points and links are paths we take to travel between the nodes. The process of
acquiring spatial knowledge involves continuously adding new nodes to the existing nodelink framework (King & Golledge, 1978).
Shopping
Despite the opposed views, the objective of this
thesis is not to take a stand on whether nodes or
links should be the basic feature that organized
Home
Work
other spatial elements into a mental layout. Instead
this thesis acknowledges that both of them are
Shopping
undeniably the prime elements that constitute the
Home
Work
basic framework of our mental representation of the
space. Thus the primary focus of this thesis is to
examine how both of these two basic elements are
remembered by a large number of city dwellers.
Sopping
The Collective Images of the City
Each resident has his/her own experience with
some parts of the city, so it seems that each
individual possesses his/her own mental map of the
city. The information we perceived from the
Work
environment
- -.
.
Figure 1.3 Golledge's anchor point theory
of spatial knowledge acquisition.
Source: Golledge, 1976
may be different
depending on
individual preferences, but what we remember is
deeply associated with the mechanism in our brain
From Spatial Perception To The Collective Images Of The City
I
II
that is in charge of understanding spatial structure, which is similar among people. Thus,
there seems to be substantial agreement among residents of the same urban neighborhood
with regard to the environment they share experience with (Lynch 1964). It is this group
consensus, represented by a collective mental map of the city, that greatly interests city
planners who wish to understand how different environmental elements facilitate or hinder
the formation of clear mental maps, and eventually design a better urban environment that
supports the uniqueness of human spatial perception.
Previous studies also investigated whether different social and economic variables might
result in disparities in the mental images between different groups of people. In one study,
Golledge (R. G. Golledge & Spector, 1978) concluded that access to an auto and amount of
time spent in the city are obvious indirect contributors to ability to accurately locate sample
places. Other studies also investigated the correlations between length of residency, where
people work or live, social classes, income and degree of familiarity (Evans, 1980).
Practical Applications
Lynch's imageability study has profoundly influenced the later practice of regional
planning and city design. The most recognizable contribution of studying the mental maps of
the city is to design a way-finding-friendly, vivid urban environment that can evoke strong
mental images. Cities with homogeneous views are monochromatic, less exciting and
sometimes frustrating. A legible city is most likely to be remembered and appreciated by its
dwellers. For example, people can easily identify a path by linking one node with another,
recall a landmark by locating it alongside a path or define a district using an edge as its
boundary. Downtown Boston, for instance, has a clear mental map. People in Boston usually
refer to Commonwealth Avenue as the path that links the Harvard Bridge and the Boston
Common, and both of them are frequently identified as major landmarks. They refer to
Beacon Hill as the district that is located next to Boston Common, and that stops at the
Charles River.
Lynch (1964) also concluded that a highly imageable city would be one whose "districts"
or "landmarks" or "pathways" are easily identifiable and are easily grouped into an overall
coherent pattern. However, studies following Lynch overemphasized the importance of a
regular city structure to an extent that they proposed cities and districts with regular shapes
12 1 Mapping Urban Perception: How Do We Know Where We Are?
formed by parallel streets and perpendicular intersections. We all know that one effective
strategy to compose highly imageable views is to use distinctive elements in terms of size,
color and form. Regular cities might be easy to navigate in, but their repeating grid structure
might eventually evoke tiresome.
Each individual's perception of the urban environment is determined by the information
he/she obtained in terms of location, proximity, distance and area, which furthermore
influences his/her everyday behavior (Golledge 1999). People make mobility-related
decisions such as which route to take, which coffee shop to go to, and whether to drive or
walk based on their memories of the spatial relationships acquired through their everyday
activities. They tend to pay more visits to places that seem easier to get to than others. If a
place seems further away than it really is, they may tend to take the car instead of walking
there. They decide whether to make a stop at the post office based on whether they think it is
on their way to the grocery store.
As a result, another practical application of investigating mental maps of the city is to
design cites that can influence people's behavior by prudently organizing spaces, tinkering
with their size and shape to support and extend our mental penchant to comprehend physical
space (Ellard, 2010). For example, when designing a pedestrian-friendly neighborhood, one
strategy is to make it feel smaller than it really is by inserting highly recognizable
destinations within walking distance of each other. People are more likely to think the post
office is close to their home if they think of it as only a few steps from the bookstore, which
is a few steps from the supermarket, which is only a few steps from home.
The fundamental belief here is that people make living decisions based on the perceived
urban environment. And if by studying the mental images the citizens have of their own city,
knowing their preferences for, perception of, and anticipation towards various urban images,
city planners can design improved environments that induce better decision making (Garling
& Golledge 1989).
Chapter 3
METHODOLOGY
Those were lengthy interviews, normally lasting about one and a half hours,
but almost always attended with great interest by the subjects.
-Kevin Lynch (1964)
One major challenge that researchers face when investigating people's mental images of
the city is how to externalize an individual's spatial knowledge of a familiar environment.
This chapter discusses the conventional methods that were used in urban perception studies
and the problems and concerns they raised. Alternative methods are proposed, in particular,
the use of online visual surveys with geo-tagged images to collect quantitative data of
perceived urban environment.
Spatial Familiarity Evaluation
The commonly accepted theory on constructing mental maps is that their basic units are
places that are "most familiar" to the great proportion of residents (Gale 1990). The concept
of spatial familiarity is, by definition, the ability to be able to identify places that one has
physically visited or encountered by other means, such as looking at a photo, trying to recall
its name or seeing it on the television (Gale 1990). However, it is difficult to measure spatial
familiarity, because it is an internal measurement of our knowledge of the space. That people
sometimes have a hard time recalling the name of a place they visit frequently doesn't
necessarily suggest that they are not familiar with that place.
Gale (1990) states that familiarity can be assessed or inferred; his study attempts to
decompose spatial familiarity as an operational variable in the models of spatial knowledge
and behavior. He designs experiments to test different indicators of spatial familiarity:
location knowledge, visual recognition, name identification and interaction frequency.
According to Gale, the most direct indicators of spatial familiarity is the "sense of locational
knowledge," that is, being able to clearly point out where this place is verbally or by pointing
to a map (Gale, Golledge, Halperin, & Couclelis, 1990). The second spatial familiarity
indicator is visual recognition, which is the ability to remember "having seen it before" when
presenting a visual representation of the place. The other familiarity indicators are title
recognition and interaction frequency. Gale (R. G. Golledge, 1999) was able to conclude that
14 1 Mapping Urban Perception: How Do We Know Where We Are?
the degree of a person's familiarity with a place is highly collinear with visual recognition
and locational accuracy.
Sketch Maps
To measure and analyze spatial knowledge, despite the different understandings of its
internal representation, the most common practice to externalize the stored spatial
information is to have subjects produce hand-drawn sketch maps. Subjects are asked to draw
a map of a specific urban area based on their memories. Some researchers believe that the
maps we draw are the best evidence of the spatial relations in our minds.
Figure 2.1 Lynch, K. (1964). Composite map of Boston.
Retrieved from http://bostonography.com
The statement is to some extent accurate and has received certain theoretical validation.
In order to retrieve spatial knowledge, people first need to image and reconstruct spaces in
their minds. The reconstructed internal representation has some map-like qualities and is
primarily embedded with the topological relationships of the physical spaces. Sketch maps
can reveal some general spatial cognition measurement such as the relative location of places,
shapes of elements, or even cognitive distance between places. However, the results of sketch
maps are very hard to quantify and compare. This is mainly because individuals' drawing
abilities vary dramatically. A person drawing a public square much larger than it really is
doesn't necessarily indicate that the person bears a stronger image of the square. Some
Methodology 1 15
researchers treat sketch maps with great scrutiny, using measurements retrieved from the
sketch map to indicate level of perceived information. Golledge has argued that sketch map
data should be treated either descriptively or analyzed within an ordinal scale (Evans, 1980).
Surveys and Interviews
Other commonly used methods to solicit response from subjects to analyze how they
comprehend space are verbal interviews, questionnaires and cognition tasks. In Lynch's
study, he conducted five years of interviews and surveys among residents of three different
cities. Examples of the questions he asked are: what symbolizes the word "Boston"? Please
give me complete and explicit directions for the trip that you normally take going from home
to where you work. What elements of central Boston do you think are most distinctive?
(Lynch, 1964). Approximately 100 people from each city participated in the interview, which
lasted normally one hour and half and was entirely recorded.
Golledge and Spector's (1978) study interviewed a sample of 151 local people to elicit
information concerning the spatial structure of the mental maps of Columbus, OH. People
were asked to compare a sample of distance relationship between forty-nine well-known
places (Figure 2.2a). The experiment attempted to predict knowledge of interpoint distance
relationships between those places as a function of where and how extensively individuals
travel about the city and (by inference) their need to store and organize spatial information
(R. G. Golledge & Spector, 1978). The result was constructed as a contour map representing
the imaged relative location of places in Columbus (Figure 2.2b).
Figure2.2a Location of cues in
Golledge's Columbus study
Figure2.2b Contours of Location Errors in
Golledge's Columbus study
16 1 Mapping Urban Perception: How Do We Know Where We Are?
Problems and Concerns
The top concern when studying images of a city processed by its citizens is the lack of an
efficient method to collect urban perception data. Sketch maps are inaccurate during
quantitative analysis and carry significant amount of noise due to individuals' drawing
abilities. Interviews require a considerable amount of time to interact with participants.
Previous studies frequently attempted to generalize findings about the mental images of a
large urban environment based on observations of a small fraction of its inhabitants.
Another concern is the degree of schematization that applied to the overall environment.
The collective urban perception and knowledge are complex structures that derived from the
formless fragmented urban cities. Yet when generating the cognitive representation of large
cities, researchers frequently adopt methods discussed above that simplify, structure, and
stabilize the flux of environmental experience (Appleyard, 1973). It is reasonable that,
according to Appleyard (1973), people have to schematize drastically if they wish to
understand the overall comprehension of urban structure, primarily due to the difficulty of
collecting a vast amount of fragmented urban experience and the lack of adequate techniques
to quantify cognitive data
The urban cities are inhabited by millions of people; however the number subjects
involved in previous studies are extremely limited. To revolutionize the investigation of
collective mental images of a city, there is a strong need for an efficient, automated and
scalable method to collect and analyze urban perception data, bridging urban perception with
other quantitative measurement in urban studies with respect to social, demographic,
economic and cultural context of cities.
Visual Surveys with Geo-tagged Images
In 2010, a pioneering study named "Place Pulse" conducted by the Micro Connection
group at the MIT Media Lab used crowd-sourced visual surveys to solicit responses towards
streetview images from different cities (Salesses, Schechtner, & Hidalgo, 2013). The group
built a website to ask people to vote on how they feel about a variety of urban images derived
from Google's streetview engine. Visitors of the website were presented with two streetviews
from two different cities. They were then asked to vote based on the questions such as "in
Methodology 1
which city do you feel safer," "which place looks more beautiful," "which place looks
wealthier," "which place looks more depressing" etc. The "Place Pulse" has successfully
collected more than 500,000 votes, based on which they conducted analysis on urban
perception and characterized the inequalities among cities from all over the world (e.g., New
York City, Boston, Salzburg, London, Tokyo, Bangkok etc.). This study concluded that
online images could be used to create reproducible quantitative measures of urban perception
among cities images (Salesses, Schechtner, & Hidalgo, 2013).
By using online visual surveys, the availability of participants has exponentially
increased. It also enables us to capture great details about every corner of the urban
environment. Researchers don't have to conduct hour-long interviews with selected subjects,
who have to make time and arrange visits to the institute. By carefully designing the user
interface of the survey website, implementing interactive features, providing instructions that
are easy to follow, the researchers enable participants to perform required actions without
face-to-face guidance.
PLACE PULSE
s1.s Clicks
Ran rgs
VISoIn
Data
experiments
papers
About
Which place looks more depressing?
4
For this question: 90,791 cllcxs collected
Goal: 500.000 cic ks
SEE REAL-TIME RANKINGS
CUCKS
TREND
1
Rio De Janeiro
6293
S4
Singapore
4286
.
2
Belo Horizonte
3198
55
Atlanta
6529
3
Gaborone
1135
56
Washington DC
1549
RANK
CITY
CLICKS
TREND
RANK
CITY
Figure 2.3 Place Pulse
Retrieved from http://pulse.media.mit.edu/
17
Chapter 4
THE EXPERIMENT
We created UrbanExplorer because we want to make a game about [your
own city]. We want this game to help you know your city better.
-UrbanExplorer
The objective of this experiment is to evaluate the familiarity of different places in a
small urban area using an online visual survey and geo-tagged streetview images.
Participants were shown random streetviews from their familiar neighborhood and asked to
guess where this place was by pointing to a map. Thus, the person's familiarity with a
specific image is assessed by two joint indicators: whether it can be recalled by its visual
image and whether it can be located accurately on a map.
Tool Design
The experimental
idea was inspired by the popular online game GeoGuessr
(http://geoguessr.com/), a geography game that takes people to random places around the
world and tests their ability to recognize the surroundings. The website designed for this
study is named Urbanexplorer (http://uexplorer.meteor.com/) and designed as a game to help
people explore their familiar urban neighborhood (Figure 3.1).
Urbanexplorer takes advantage of thousands of geo-tagged streetview images retrieved
from the Google streetview API. Urbanexplorer encourages participants to play the game by
asking the challenging question: how well do you know Cambridge? Guesses are scored
based on the time taken to place the guess and the distance between the guessed location and
the real location of the image. This way, people are encouraged to place their guesses as
accurately as possible without hesitating for too long. On average, participants took 2-3
guesses to gain proficiency at the required actions, such as clicking in the map, looking for
the precise locations and moving on to the next guess. Thus, in the analysis, the first two
guesses placed by each participant are ignored. Each individual's familiarity with a given
place is a function of reaction time and guessed distance to answer. The overall familiarity of
the place can be inferred from the collective actions and responses of all players. Streetview
images are shown to participants in random order. Once participants have placed a guess, the
20 1 Mapping Urban Perception: How Do We Know Where We Are?
real location of the streetview is revealed. All the guesses are stored in the database with a set
of attributes (Table 3.1).
Place id:
Each streetview is given an unique id for easy retrieving
Time spent to guess: Time is calculated from the moment a streetview is
shown to the moment the person clicks the guess button
Location of streetview:
Longitude and Latitude of the streetview
Location of guess: Longitude and Latitude of the guessed location
Distance to answer: Distance are calculated between the two locations above
User information:
Information of the user matched with his/her survey
Time of creation: Time when the guess is made
Score:
Score is a function of distance and time spent based on a
scale of I to 2500
Table 3.1 Attributes associated with each guess
AEURBAN
EXPLORER
0
Pce rrailrkrto make. a uss X
Figure 3.1 The User Interface of UrbanExplorer
Retrived from http://uexplorer.mit.edu
cr
ae
ntm
n
The Experiment 1 21
Survey
In order to collect data on the demographics of the participants, as well as exclude
answers of participants who have never been to Cambridge or the studied area, the website
asks them to answer a three-question survey after they have placed three guesses looking at
images of Harvard Square. The survey questions are:
1. How do you know Cambridge? I ... Cambridge (multi-selection).
A. live/lived in
B. work/worked in
C. live/lived and work/worked in
D. visited
E. have never been to
2. How old are you?
A. less than 25
B. 25-34
C. 35-44
D. 45.54
E. older than 55
3. How many times have you visited Harvard Square?
A. less than 3 times
B. 3-10 times
C. more than 10 times
Site
The urban area that was chosen for this study is Harvard Square in Cambridge, MA.
Located in the center of Cambridge, adjacent to Harvard University, Harvard Square is an
area frequently visited by tourists and local residents. It consists of a variety of local
attractions, such as restaurants, salons, clothing shops, bookstores, coffee shops, florists and
hotels. The MBTA stop "Harvard Square" is a major transportation hub that links subway,
bus and taxi stops, transporting thousands of people to and from this area every day. Most of
the people who live in Cambridge or adjacent neighborhoods have visited Harvard Square in
the past.
22 1 Mapping Urban Perception: How Do We Know Where We Are?
The area that was chosen to study covers the center of Harvard Square, extends north to
Cambridge Common and south to the river. The land-use types of the studied area are mainly
commercial, transportation and educational. The road system of this area is extremely
complex and irregular. Most of the roads don't follow the square block system, and most of
the major intersections are not perpendicular. The highly visited public places, various land
uses and irregular structure render it an ideal location for our study.
Figure 3.2 Arial Photo of Harvard Square
Retrieved from http://maps.google.com
Geo-tagged Streetview Images
For this study 190 geo-tagged streetviews around Harvard Square were manually
selected, along with 50 streetviews of other popular places in Cambridge. Each streetview
has a predefined point of view, so that a participant cannot look around. The purpose of
including images from outside the Harvard Square is to prevent participants from only
placing guesses in that area. The images around Harvard Square were categorized into two
groups representing nodes and links. The nodes are images of intersections, public squares
and single buildings that, according to Lynch's five elements and Golledge's anchor point
theory, are usually referred to as spatial reference points. The links are images of the street
section between the nodes. For each node, 3 to 4 images are selected looking at different
points of view. For each link, 2 images are selected looking at different directions.
All 240 images are randomly picked and shown to people by the program. Each image
will only be shown once to the same participant. On average, each participant placed 10
guesses before quitting the game.
The Experiment 1 23
Figure 3.3 Location of all street views selected around Harvard Square
Figure 3.4 Samples images of the link
Figure 3.5 Sample images of the node
24 1 Mapping Urban Perception: How Do We Know Where We Are?
Participants
The survey was sent primarily to people who were associated with universities. Most of
them were students or faculty members. In all, 394 people participated in this study. Most of
the participants were people who currently live and/or work in Cambridge. Among these,
68.0% of the participants stated they currently live or work in Cambridge, or have done so in
the past; 79.4% of the subjects claimed to have visited the studied area more than 10 times in
the past, and another 15.5% claimed to have visited the studied area 3-10 times; and 84.0% of
the subjects are under 35 years old.
Figure 3.8 Ages of all participants
Figure 3.6 Visit frequencies of all participants
The Experiment 1 25
Figure 3.7 Modes of interactions of all participants
Algorithm
For each image, a score indicating how familiar people are with it was calculated based
on all guesses. The score is the sum of the percentage of guesses that were made within 50
meters, the percentage of guesses made between 50 to 100 meters multiplied by 0.75 and the
percentage of guesses made between 100 to 150 meters multiplied by 0.5. The range of the
score is 0 to 1, 1 indicating high familiarity. An image will receive a score 1 if all the guesses
of it were made within 50 meters. The calculation is shown below.
Q,
- Number of guesses made within 50 meters of the answer
Q2 - Number of guesses made between 50 to 100 meters to the answer
Q3 - Number of guesses made between 100 to 150 meters to the answer
Q
- Total number of valid guesses
S = (QI
x
1 + Q2 x 0.75 + Q3x 0.5) /
Q
Chapter 5
RESULTS AND DISCUSSIONS
The creation of the environmental image is a two-way process between
observer and observed. What he sees is based on exterior form, but how he
interprets and organizes this, and how he directs his attention, in its turn
affects what he sees. The human organism is highly adaptable and flexible,
and different groups may have widely different images of the same outer
reality.
-Kevin Lynch, The Image Of The City
In total, 4216 guesses were made, of which 3617 were made by people who filled out the
survey, and 2851 guesses were made to guess the 190 streetviews in Harvard Square. On
average each image received 15 guesses.
Figure 5.1a Locations of the 15 most recognized streetviews
28 1 Mapping Urban Perception: How Do We Know Where We Are?
Most Familiar and Least Familiar
Not surprisingly, the places with the highest scores are primarily public spaces at the
center of the studied area, such as Harvard Square, Brattle Square and Winthrop Square.
Certain stores at the center of Harvard Square (e.g., Harvard Coop) and along Massachusetts
Avenue (e.g., Qdoba Mexican Grill) also received high scores. In fact, the place that is most
recognized is the Qdoba Mexican Grill along Massachusetts Avenue, with a score of 0.847.
The links with the highest score are streets within one block to the center (e.g., JFK Street
and Massachusetts Avenue). This finding indicates that the degree of interaction and the
proximity to center is highly correlated with the probability of being recognized by most
people.
Figure 5.ab The 15 most recognized streetviews
The observation also suggests that the irregularity of urban structure doesn't prevent
forming of strong mental images. Most of the highly recognized places are in the center of
Results and Discussion 1 29
the studied area, where road system is extremely complex. The shapes of the highly
recognized squares are triangular, and the roads are circled. This finding challenges Lynch's
statement that a regular city structure is more likely to evoke strong mental images.
The places with the lowest scores are primarily school or residential buildings with no
eye-catching signs or distinct features, as well as streets that don't have much activity (e.g.,
Garden Street, Bow Street and Arrow Street). It is very obvious that the least recognized
places are all located away from the center of the studied area. Some findings that are worth
mention include:
1. The second least recognized streetview with a score of 0.022 is the building of the John
F. Kennedy School, located at the intersection of JFK Street and Eliot Street. JFK Street is
the busiest street in Harvard Square with heavy traffic and frequent congestion. The fact that
the building at one busy intersection is very unfamiliar to people indicates that frequently
passing by a place does not necessarily evoke strong images.
Figure 5.2a Locations of the 15 least recognized streetviews
30 1 Mapping Urban Perception: How Do We Know Where We Are?
2. The sixth least recognized streetview with a score of 0.074 looks at the back of the
Harvard University Office Building, which takes up the entire block. However, the eleventh
most recognized streetviews looks at the front of the same building, where large red windows
decorate the three-story podium in the center of the view. This observation suggests that
large-size buildings do not necessarily evoke strong mental images, because pedestrians are
more likely to look at features visible at eye level.
Figure 5.2 b The 15 least recognized streetviews
Results and Discussion 1 31
Links and Nodes
Of all the answers placed within 100 meters of the correct location, 30% of the answers
of the nodes were concentrated within 20 meters of the true location, whereas only 15% of
the answers of the links were within 20 meters. This indicates that if people recognize the
image, they tend to place the marker more accurately (within 20 meters to the real location)
when the image they are guessing is a node. On the other hand, guesses for links tend to be
less accurate.
Figure 5.3 Left: All guesses for links. Right: All guesses for nodes
Figure 5.4 Histogram of guessed distances
32 1 Mapping Urban Perception: How Do We Know Where We Are?
Figure 5.5a shows the scores of all nodes. The size of the ellipses indicates the score. The
nodes more familiar to people are located along Massachusetts Avenue and JFK Street,
which are the two busiest streets. Most of the nodes along these two streets point to stores or
shops along the street. Along Massachusetts Avenue, the more familiar views are mostly
towards the south side of the street. On the other hand, among the nodes along Mt. Auburn
Street, the more familiar views are all towards the south. A magnetic field seems to exit
between these two parallel streets that draws more attention. This is the area that has many
restaurants and cafes.
Figure 5.5a Score of nodes
Results and Discussion
|
33
Figure 5.6b shows the score of all links. The more familiar streets are Massachusetts
Avenue and JFK Street. A pattern can be observed among the streets between Massachusetts
Avenue and Mt. Auburn Street. The closer the street is to the center, the more familiar it is to
most people. This again verified the earlier observation that proximity to center has strong
correlation with the degree of familiarity.
Figure 5.6b Score of links
An interesting finding was observed when conducting high-resolution analysis of every
angle of this urban area. This experiment asked participants to guess streetviews without
rotating the view, which allows us to compare the familiarities of different viewpoints at the
same geographic location. At most locations, the familiarities of different viewpoints vary to
some extent. In fact, at certain locations, the familiarities between different viewpoints are
dramatically different. For example, the shop next to the Qdoba Mexican Grill-the most
34 1 Mapping Urban Perception: How Do We Know Where We Are?
recognized
place-is
very
poorly
recognized,
although these two places are only 10 meters apart.
At viewpoint A, the Qdoba Mexican Gill, 8 out of
12 people made a guess within 20 meters. In
contrast, at viewpoint B, Zinnia Jewelry, only 2 out
of 14 people made a guess within 20 meters, and 10
people didn't even guess within 800 meters, which
suggested that they couldn't recognize this place at
all. There are many possible explanations. Qdoba is
a fast food restaurant, which people frequently visit,
whereas Zinnia is a small jewelry shop that does not
attract many people. Another possible explanation is
Figure 5.7 Two different viewpoints at the
intersection of Massachusetts Avenue and
Linden Street
that the church tower at the very far end of image A
provides a hint to the location of this streetview.
Figure 5.7 Left: Poorly recognized viewpoint. Right: Highly recognized viewpoint.
Results and Discussion
|
35
Knowledge Acquisition
Figure
5.8
shows
how
spatial
knowledge is acquired when more visits
are paid. From top to bottom, images show
the familiarity of places among people
who have visited this area less than three
times, between three and ten times and
more than ten times. When people visit
once or twice, spatial knowledge of the
activity center is first obtained. When
more visits are paid, people start to gain
knowledge of the spaces that are spatially
close to the center, including the part of
the streets that connect to the center of this
area, and intersections that are one or two
blocks away from the center. Finally,
when people have visited the place more
than ten times, the structure of their spatial
knowledge
starts
to
appear.
This
observation seems to favor the anchorpoint theory, which proposes that the
spaces in our mind are formed as a linkednode structure. However, the anchor-point
theory is mainly concerned with each
individual's spatial knowledge. It might
not apply to the collective mental maps of
the city. What can be observed here is the
spatial knowledge system
seems have
some kind of hierarchy embedded, that it
Figure 5.8 Above: Score of people visited less than 3
times. Middle: Score of people visited 3-10 times.
Below: Score of people visited than 10 times.
can only grows from the center.
Results and Discussion 1 36
Modes of Interaction
Figure 5.8
interactions
shows how modes of
affect
spatial
familiarity.
From top to bottom, images show the
familiarity of places among people who
only live in Cambridge, only work in
Cambridge,
and
Cambridge
at
live
the
and
same
work
in
time.
In
comparison, the familiarities among people
who only live in Cambridge extend more
towards the
surrounding area,
where
mostly residential districts, rather than
concentrated in the center activity area.
The familiarities among people who only
work in Cambridge concentrate along the
busy streets and center activity area, not so
much on the parallel streets in between
Massachusetts Avenue and Mt Auburn
Street. The familiarities among people who
work and live in Cambridge at the same
time are confined to the middle portion of
the studied area framed by Massachusetts
Avenue and Mt Auburn Street, including
the parallel streets in between. All the
places in the middle area are very well
recognized by people who like and work in
Cambridge. The observation suggests that
the modes of interactions can affect the
distribution of familiar palces. In general,
Figure 5.Error! No text of specified style in document..9
Above: Score of people live in Cambridge. Middle:
Score of people work in Cambridge.
Below: Score of people live and work in Cambridge
people know places better if they work
and live in the area at the same time.
Results and Discussion 1 37
Discussion
Many of the results that are discussed above are simply observations from the data this
study collected, and the results confirmed statements in some previous studies, such as that
familiarity is correlated to the proximity to center, the degree of interaction determines the
level of familiarity and modes of interaction affect distribution of familiar places. There is
also interesting findings that were only observable when conducting research using this finegrained approach. Findings, such as that different viewpoints of the same location might
induce extremely high and low familiarity, two viewpoints toward different directions at the
same street might be highly and poorly recognized, can only be observed when we are able to
ask a large number of people to guess every corner of the streets in the studied area.
We were also able to observe differences among various groups of people in respect to
level and distribution of familiarity. Comparison was made among people with different
numbers of visits, modes of interaction and ages. More studies can be done to further analyze
different cultural, economic and class factors that affect familiarity, by adding questions to
the survey such as "what is your occupation," "what are your primary activities when visiting
this area," and "what modes of transportation do you usually choose when coming to this
area."
Further investigation can be conducted to analyze the semantics between familiar and
unfamiliar places, such as whether a high familiarity place will induce higher familiarities
among views that contain such a place; whether the familiarity of different directions of one
link is determined by the nodes that it connects; and whether some type of magnetic field can
be calculated at certain areas. We believe that views are connected, and the familiarities of
places in the same area are also connected, the relationship between which can be calculated
through network analysis.
Urban perception is related to human mobility behavior. Future studies in this field can
use the method demonstrated in this thesis to verify mobility data, to determine the causal
relationships between these two factors. Some of the observations in the study can also be
verified when more data related to how people move around in the studied area are collected.
Chapter 6
VISUALIZATIONS
What would happen if, when saying that some image is human-made, you
were increasing instead of decreasing its claim to truth?
-Bruno Latour,Iconoclash
With 394 participants placing on average 10 guesses, we were able to build interactive
visualizations to represent how well places are remembered on a map. The visualization
below presents the collective memory of Harvard Square by the study participants. The size
of the ellipse represents the degree of familiarity. Instead of schematized approach, this study
demonstrates the use of computer algorithm and mapping techniques to process and calculate
data collected from a large number of people, and represents the results using interactive data
visualizations. The familiarities of each view are automatically calculated by querying all the
guesses stored in the database. This visualization is also immediately updated when new
guesses are made.
Figure 6.1 Interactive data visualization showing the score of all streetviews
40 1 Mapping Urban Perception: How Do We Know Where We Are?
Comparison can be easily made between different data sources. Foursquare is a locationbased social networking website for mobile devices. Users "check-in" at venues using a
mobile device by selecting from a list of venues the application locates nearby (Wikipedia).
Foursquare data is a direct indicator of the how alive an area is. Below is a map generated by
overlapping Foursquare check-ins with the score of all streetviews. The size of the blue circle
represents the number of check-ins at that location. Trends can be observed that high checkin locations are always correlated with highly recognized streetviews, especially at the center
of studied area (e.g., Harvard Square, Massachusetts Avenue, JFK Street etc.)
Figure 6.2 Map overlapping Foursqure check-ins with the score of streetviews
An interactive data visualization platform was built to study and visualize the collected
data. A user is able to browse between different aggregations of data, for example,
differentiating nodes and links, making comparison between people among different numbers
of visits, modes of interactions and ages. It also enables researchers to look at each streetview
at high resolution, by showing the location, ranking, distances and user profile of all the
guesses.
Visualizations 1 41
U,
4 O 1; -.
WW
127.0.0.1
Figure 6.3a Interactive data visualization platform visualizing nodes vs. links
11700 lJlb00W700~
4-
~*
Interactive data visualization visus
poeple between 25-34 years old
42 1 Mapping Urban Perception: How Do We Know Where We Are?
I
~-
e
1~7.O.0.2/~rn
~
%I
Visualizations
|
43
This thesis collected the spatial knowledge around Harvard Square of 394 people. It
painted the collective memory of the city using people's memory, highlighting the most
remembered and least remembered sections of the cityscape. This is a first attempt made to
paint a drawing of the collective memory of the city using literally people's memory. We
wish to collect all memories of the city and create a similar painting of cities of all the
continents.
Figure 6.5 The collective memory of the city
Chapter 7
CONCLUSIONS
There is a strong need to create quantitative bridges that can help us link
urban perception with other social, political, economic and cultural aspects
of cities.
-
Salesses, Schechtner, & Hidalgo, The Collective image of the City
The mental map of an urban environment is crucial for the daily activities of thousands of
city-dwellers. Yet until now, the availability of data about urban perception has been limited,
and so has our ability to collect a large amount of fine-grained data representing how people
comprehend their familiar urban environment. In this thesis, we presented a method to
automatically collect people's responses to measure spatial familiarity of two basic elements:
link and node. Using a website to conduct a visual survey, we invited 394 participants to
guess the location of 240 geo-tagged street views captured from a familiar urban
environment. We then used quantitative calculation and data visualization techniques to
measure how well different places are remembered by people.
This analysis confirms some of the earlier statements that were made. It concludes that
the familiarity is a function of proximity to center and degree of interaction. Places closer to
the center of activity are more likely to be remembered. However, proximity alone does not
result in high familiarity. Places along a busy street can be poorly recognized if most of the
people had limited or no interaction with such a place. The distinctive features of the building
can evoke high imageability if the feature is constantly visible from a pedestrian's viewpoint.
The unique findings of this thesis based on the fine-grained data are related to how
familiarity can be dramatically different at the same geo-location. Google's geo-tagged
images that this thesis collected enabled us to survey every corner and every angle of the
studied area. We were able to observe that two viewpoints of the same location have
dramatically different familiarities. One is extremely highly remembered (8 out of 12 people
guessed within 20 meters), and the other is very poorly remembered (2 out of 13 people
guessed within 20 meters). Views of the same building from different angles also result in
46 1 Mapping Urban Perception: How Do We Know Where We Are?
diverse familiarity: the back view of an office building along a busy street is the
6
h
least
remembered view among 190 streetviews in that area, whereas the front view of the building
is the 11h most remembered view. These findings suggest that when general distribution of
familiarities in a small area can be observed as highest at the center of activity and declining
towards the edge, differences in viewpoint can still be dramatic. People's spatial knowledge
is complex and fragmented. The findings also suggest that the irregularity of urban structure
doesn't prevent forming of strong mental images, which challenges Lynch's statement that a
regular city structure is more likely to evoke strong mental images.
This thesis also observed that spatial knowledge was acquired throughout time. With
limited visits, people are only able to remember some spots that they have limited interaction
with. The more visits paid, the more spots will be added to their memory starting from the
vicinity of the center. This to some extent in alignment with with Golledge's (R. G. Golledge
& Spector, 1978) anchor-point theory. This thesis also observed that modes of interactions
can affect distribution and level of familiarity.
This thesis also explores how cartographic techniques can be used to map the virtual
memory of the city that its inhabitants and those of other neighborhoods have of it. By
analyzing data collected using an interactive data visualization platform, we are able to
conduct fine-grained statistical analysis at each streetview and among different groups of
participants. We are also able to efficiently conduct comparison by aggregating data
differently or overlapping with other sources of urban data in demographic, economic,
cultural, social media context.
This thesis presents an innovative approach using web-based visual surveys to conduct
studies of urban spatial perception in an efficient, automatic and scalable way. Using online
surveys and geo-tagged images, we are able to collect a large amount of fine-grained data
from a large number of participants. Compared to conventional methods such as interviews
and face-to-face surveys, online surveys eliminated the time required to interact with each
participant, exponentially enlarging the number of participants that can be involved.
Clearly, there are limitations. One concern is the lack of control over the demographics of
the participants. People who are willing to take the survey are more likely to be medium or
Conclusions 1 47
higher income groups, who have access to the Internet and are willing to voluntarily
participate in academic research. Another problem is that there is no means to filter irrelevant
data. People might carelessly click around without paying attention to the instructions. Future
improvements can be made using machine learning to control and balance demographics of
the users, and designing better user interface to eliminate user errors.
This thesis concludes that the method presented here is able to rapidly elicit spatial
knowledge among a large number of city dwellers, and allow researchers to conduct
quantitative data analysis. Moreover, this thesis proposes that the method can be used beyond
the familiarity evaluation of urban images. It can be adopted in future investigations
including spatial knowledge acquisition in an unfamiliar neighborhood, the semantic
relationships between spatial knowledge, and results of perceptional knowledge on human
mobility behavior.
BIBLIOGRAPHY
Abrams, J., & Hall, P. (2006). Else/where : mapping new cartographiesof networks and
territories/ Janet Abrams + PeterHall, editors. Minneapolis, MN : University of
Minnesota Design Institute: Distributed by University of Minnesota Press, c2006.
Appleyard, D. (1973). Notes on urban perception and knowledge. Image and Environment.
Cognitive Mapping and SpatialBehavior, 109-114.
Bahrick, H. P. (1983). The cognitive map of a city: fifty years of learning and memory.
Psychology of Learning & Motivation, 17, 125-163. doi: 10.1016/S0079-7421(08)600985
Chombart de Lauwe, P. H. (1965). Paris,essais de sociologie, 1952-1964. Paris:
Ouvrieres.
Editions
Dodge, M., Kitchin, R., & Perkins, C. (2011). Cognitive Maps and Spatial Behaviour:
Process and Products. Map Reader, 312.
Ellard, C. (2010). You Are Here: Why We Can FindOur Way to the Moon, but Get Lost in
the Mall (1 edition.). New York: Anchor.
Evans, G. W. (1980). Environmental cognition. PsychologicalBulletin, 88(2), 259-287.
doi:http://dx.doi.org.libproxy.mit.edu/1 0.1037/0033-2909.88.2.259
Foursquare. (2014, May 25). Wikipedia. Retrieved May 26, 2014, from
http://en.wikipedia.org/wiki/Foursquare
Gale, N., Golledge, R. G., Halperin, W. C., & Couclelis, H. (1990). Exploring Spatial
Familiarity*. The ProfessionalGeographer,42(3), 299-313. doi:10. 11 1/j.00330124.1990.00299.x
Garling, T., & Golledge, R. G. (1989). Environmental perception and cognition. In E. H.
Zube & G. T. Moore (Eds.), Advances in Environment, Behavior, and Design, Volume 2
(pp. 203-236). New York: Plenum Press.
Golledge, R. G. (Ed.). (1999). Wayfinding behavior: cognitive mapping and other spatial
processes. Baltimore: Johns Hopkins University Press.
Golledge, R. G., & Spector, A. N. (1978). Comprehending the Urban Environment: Theory
and Practice. GeographicalAnalysis, 10(4), 403-426. doi:10. 111 /j.15384632.1978.tb00667.x
Golledge, R. G., & Stimson. (1997). Spatial behavior: a geographicperspective. New York:
Guilford Press.
Herzog, T., Kaplan, Stephen, Kaplan, Rachel. (1976). The Prediction of Preference for
Familiar Urban Places. Environment and Behavior, 8(4), 627-645.
50
| Mapping Urban Perception:
How Do We Know Where We Are?
King, L. J., & Golledge, R. G. (1978). Cities, space, and behavior: the elements of urban
geography.Englewood Cliffs, N.J.: Prentice-Hall.
Kitchin, R., & Freundschuh, S. (Eds.). (2000). Cognitive mapping: past,present, andfuture.
London; New York: Routledge.
Lynch, K. (1964). The image of the city (1st M.I.T. Press pbk. ed.). Cambridge, Mass: M.I.T.
Press.
Nasar, J. L. (1998). The evaluative image of the city. Thousand Oaks, CA: Sage Publications.
Place recognition and Wayfinding: Making Sense of Space. (n.d.).
Salesses, P., Schechtner, K., & Hidalgo, C. A. (2013). The Collaborative Image of The City:
Mapping the Inequality of Urban Perception. PLoS ONE, 8(7), 1-12.
doi: 10.1371/journal.pone.0068400